A comprehensive method for sensor fault detection and signal reconstruction in structural health monitoring systems for super-tall buildings
Year: 2025
Type: article
Abstract: Structural Health Monitoring (SHM) systems are crucial for ensuring the integrity and safety of large-scale infrastructures. However, they are often compromised by sensor faults, which can result in false alarms or data gaps. In this study, a new comprehensive approach is developed that combines Deep Signal Anomaly Detection (DSAD), One-Dimensional Convolutional Neural Networks (1D-CNNs), and Support Vector Regression (SVR) to effectively detect, classify, and reconstruct faulty sensor data in SHM systems. Using acceleration responses from the Milad Tower in Iran as a case study, the approach begins with DSAD for anomaly detection, where the threshold is determined by an Isolation Forest. The subsequent classification of fault types is performed using 1D-CNNs, optimized through Observer-Teacher-Learner-Based Optimization (OTLBO) for enhanced accuracy. Feature extraction is achieved using Wavelet Transform (WT) and Multivariate Empirical Mode Decomposition (MEMD), which facilitate a robust representation of time and time-frequency features. Faulty signals are reconstructed using SVR, with feature selection and reduction managed through Multi-Objective Observer-Teacher-Learner-Based Optimization (MOOTLBO) to minimize reconstruction errors. The results demonstrate a high fault detection accuracy of 94.04 % and effective signal reconstruction with optimized feature input, highlighting the potential of this approach to improve the reliability of SHM systems in super-tall structures. This study effectively identifies faulty signals, classifies fault types, and reconstructs them, addressing issues related to missing and faulty signals.
Source: Structures
Institutions: Kharazmi University, University of the West of England
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: hybrid
Estimated APC paid: $3,510
OpenAlex ID: W4409366244